Explainable methods for understanding deep neural networks are currently being employed for many visual tasks and provide valuable insights about their decisions. While post-hoc visual explanations offer easily understandable human cues behind neural networks’ decision-making processes, comparing their outcomes still remains challenging. Furthermore, balancing the performance-explainability trade-off could be a time-consuming process and require a deep domain knowledge. In this regard, we propose a novel auxiliary module, built upon convolutional-based encoders, which acts on the final layers of convolutional neural networks (CNNs) to learn orthogonal feature maps with a more discriminative and explainable power. This module is trained via a disentangle loss which specifically aims to decouple the object from the background in the input image. To quantitatively assess its impact on standard CNNs, and compare the quality of the resulting visual explanations, we employ metrics specifically designed for semantic segmentation tasks. These metrics rely on bounding-box annotations that may accompany image classification (or recognition) datasets, allowing us to compare both ground-truth and predicted regions. Finally, we explore the impact of various self-supervised pre-training strategies, due to their positive influence on vision tasks, and assess their effectiveness on our considered metrics.

Features Disentanglement For Explainable Convolutional Neural Networks / P. Coscia, A. Genovese, F. Scotti, V. Piuri (PROCEEDINGS - INTERNATIONAL CONFERENCE ON IMAGE PROCESSING). - In: 2024 IEEE International Conference on Image Processing (ICIP)[s.l] : IEEE, 2024 Sep 27. - ISBN 979-8-3503-4939-9. - pp. 514-520 (( convegno ICIP tenutosi a Abu Dhabi nel 2024 [10.1109/icip51287.2024.10647568].

Features Disentanglement For Explainable Convolutional Neural Networks

P. Coscia;A. Genovese;F. Scotti;V. Piuri
2024

Abstract

Explainable methods for understanding deep neural networks are currently being employed for many visual tasks and provide valuable insights about their decisions. While post-hoc visual explanations offer easily understandable human cues behind neural networks’ decision-making processes, comparing their outcomes still remains challenging. Furthermore, balancing the performance-explainability trade-off could be a time-consuming process and require a deep domain knowledge. In this regard, we propose a novel auxiliary module, built upon convolutional-based encoders, which acts on the final layers of convolutional neural networks (CNNs) to learn orthogonal feature maps with a more discriminative and explainable power. This module is trained via a disentangle loss which specifically aims to decouple the object from the background in the input image. To quantitatively assess its impact on standard CNNs, and compare the quality of the resulting visual explanations, we employ metrics specifically designed for semantic segmentation tasks. These metrics rely on bounding-box annotations that may accompany image classification (or recognition) datasets, allowing us to compare both ground-truth and predicted regions. Finally, we explore the impact of various self-supervised pre-training strategies, due to their positive influence on vision tasks, and assess their effectiveness on our considered metrics.
Explainable AI (XAI); ResNet; self-supervised learning (SSL); disentanglement
Settore INFO-01/A - Informatica
Settore IINF-05/A - Sistemi di elaborazione delle informazioni
   Edge AI Technologies for Optimised Performance Embedded Processing (EdgeAI)
   EdgeAI
   MINISTERO DELLO SVILUPPO ECONOMICO
   101097300

   SEcurity and RIghts in the CyberSpace (SERICS)
   SERICS
   MINISTERO DELL'UNIVERSITA' E DELLA RICERCA
   codice identificativo PE00000014
27-set-2024
Book Part (author)
File in questo prodotto:
File Dimensione Formato  
icip24.pdf

accesso aperto

Tipologia: Post-print, accepted manuscript ecc. (versione accettata dall'editore)
Dimensione 10.18 MB
Formato Adobe PDF
10.18 MB Adobe PDF Visualizza/Apri
Features_Disentanglement_For_Explainable_Convolutional_Neural_Networks.pdf

accesso riservato

Tipologia: Publisher's version/PDF
Dimensione 8.3 MB
Formato Adobe PDF
8.3 MB Adobe PDF   Visualizza/Apri   Richiedi una copia
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/2434/1104968
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact